Exploring the Individualized Effect of Climatic Drivers on MODIS Net Primary Productivity through an Explainable Machine Learning Framework

Round 1

Reviewer 1 Report

Identifying dominant drivers of regional vegetation productivity still is important to broaden the understanding of the spatial and temporal dynamics of terrestrial carbon fluxes.This manuscript employed a machine learning method fed with climate data to establish a prediction model with high accuracy to model the continuous MODIS NPP in the Amazon ecoregion, which determined the main drivers of NPP spatial variability in the Amazon region and revealed the individualized patterns of how each driver impacts NPP. The findings will contribute to to the ecosystem model development of carbon exchange. There are some minor flaws which need to be addressed before publication.

(1)   The part of introduction should be enhanced. Some description of the reference in the part of the discussion is suggested to be moved to the introduction.

(2)   Line163. The over 20 years data were averaged. I want to see more discussion about the reason.

(3)   XGBoost and XGBoost were used in the study, it will be better some more reason of the selection of these model and method.

（4）For the Figure 7, Figure A1，the legend is suggested to be added to show the meaning of the different color.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Overall, the significance of this work is high. Here are some minor comments that the authors could consider: 

Line 121 - Can the authors provide a one or two line summary of how consistent this dataset was with field observed data in terms of accuracy or other metrics? 

Line 181 - Were there any other tree based modes like Random Forest tried out before using XG-Boost?

The analysis only has a training and testing set but it is recommended to have a validation set while performing hyperparameter tuning. Why did the authors not consider a validation set here? 

The authors state that the hyperparmeters were partly set as follows. What does partly mean here?

What metric was used to select the hyperparameters based on test set performance? Was it the mean absolute error, or R2, or some other metric? Please provide more details here.  The hyperparameter tuning and selection section can be improve with more details. 

Line 338 - This is a strong conclusion to make based on the scope of this study. What are the authors thoughts on slowing down the rate of decrease of productivity? Please include a 1-2 line discussion. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx